Commit | Line | Data |
---|---|---|
b9170836 DJ |
1 | /* |
2 | * drivers/cpufreq/cpufreq_conservative.c | |
3 | * | |
4 | * Copyright (C) 2001 Russell King | |
5 | * (C) 2003 Venkatesh Pallipadi <venkatesh.pallipadi@intel.com>. | |
6 | * Jun Nakajima <jun.nakajima@intel.com> | |
11a80a9c | 7 | * (C) 2009 Alexander Clouter <alex@digriz.org.uk> |
b9170836 DJ |
8 | * |
9 | * This program is free software; you can redistribute it and/or modify | |
10 | * it under the terms of the GNU General Public License version 2 as | |
11 | * published by the Free Software Foundation. | |
12 | */ | |
13 | ||
14 | #include <linux/kernel.h> | |
15 | #include <linux/module.h> | |
b9170836 | 16 | #include <linux/init.h> |
b9170836 | 17 | #include <linux/cpufreq.h> |
138a0128 | 18 | #include <linux/cpu.h> |
b9170836 DJ |
19 | #include <linux/jiffies.h> |
20 | #include <linux/kernel_stat.h> | |
3fc54d37 | 21 | #include <linux/mutex.h> |
8e677ce8 AC |
22 | #include <linux/hrtimer.h> |
23 | #include <linux/tick.h> | |
24 | #include <linux/ktime.h> | |
25 | #include <linux/sched.h> | |
26 | ||
b9170836 DJ |
27 | /* |
28 | * dbs is used in this file as a shortform for demandbased switching | |
29 | * It helps to keep variable names smaller, simpler | |
30 | */ | |
31 | ||
32 | #define DEF_FREQUENCY_UP_THRESHOLD (80) | |
b9170836 | 33 | #define DEF_FREQUENCY_DOWN_THRESHOLD (20) |
b9170836 | 34 | |
18a7247d DJ |
35 | /* |
36 | * The polling frequency of this governor depends on the capability of | |
b9170836 | 37 | * the processor. Default polling frequency is 1000 times the transition |
18a7247d DJ |
38 | * latency of the processor. The governor will work on any processor with |
39 | * transition latency <= 10mS, using appropriate sampling | |
b9170836 | 40 | * rate. |
8e677ce8 AC |
41 | * For CPUs with transition latency > 10mS (mostly drivers with CPUFREQ_ETERNAL) |
42 | * this governor will not work. | |
b9170836 DJ |
43 | * All times here are in uS. |
44 | */ | |
2c906b31 | 45 | #define MIN_SAMPLING_RATE_RATIO (2) |
112124ab | 46 | |
cef9615a TR |
47 | static unsigned int min_sampling_rate; |
48 | ||
112124ab | 49 | #define LATENCY_MULTIPLIER (1000) |
cef9615a | 50 | #define MIN_LATENCY_MULTIPLIER (100) |
2c906b31 AC |
51 | #define DEF_SAMPLING_DOWN_FACTOR (1) |
52 | #define MAX_SAMPLING_DOWN_FACTOR (10) | |
1c256245 | 53 | #define TRANSITION_LATENCY_LIMIT (10 * 1000 * 1000) |
b9170836 | 54 | |
c4028958 | 55 | static void do_dbs_timer(struct work_struct *work); |
b9170836 DJ |
56 | |
57 | struct cpu_dbs_info_s { | |
8e677ce8 AC |
58 | cputime64_t prev_cpu_idle; |
59 | cputime64_t prev_cpu_wall; | |
60 | cputime64_t prev_cpu_nice; | |
18a7247d | 61 | struct cpufreq_policy *cur_policy; |
8e677ce8 | 62 | struct delayed_work work; |
18a7247d DJ |
63 | unsigned int down_skip; |
64 | unsigned int requested_freq; | |
8e677ce8 | 65 | int cpu; |
26d204af | 66 | unsigned int enable:1; |
ee88415c | 67 | /* |
68 | * percpu mutex that serializes governor limit change with | |
69 | * do_dbs_timer invocation. We do not want do_dbs_timer to run | |
70 | * when user is changing the governor or limits. | |
71 | */ | |
72 | struct mutex timer_mutex; | |
b9170836 DJ |
73 | }; |
74 | static DEFINE_PER_CPU(struct cpu_dbs_info_s, cpu_dbs_info); | |
75 | ||
76 | static unsigned int dbs_enable; /* number of CPUs using this policy */ | |
77 | ||
4ec223d0 | 78 | /* |
7d26e2d5 | 79 | * dbs_mutex protects data in dbs_tuners_ins from concurrent changes on |
ee88415c | 80 | * different CPUs. It protects dbs_enable in governor start/stop. |
4ec223d0 | 81 | */ |
9acef487 | 82 | static DEFINE_MUTEX(dbs_mutex); |
b9170836 | 83 | |
8e677ce8 AC |
84 | static struct workqueue_struct *kconservative_wq; |
85 | ||
86 | static struct dbs_tuners { | |
18a7247d DJ |
87 | unsigned int sampling_rate; |
88 | unsigned int sampling_down_factor; | |
89 | unsigned int up_threshold; | |
90 | unsigned int down_threshold; | |
91 | unsigned int ignore_nice; | |
92 | unsigned int freq_step; | |
8e677ce8 | 93 | } dbs_tuners_ins = { |
18a7247d DJ |
94 | .up_threshold = DEF_FREQUENCY_UP_THRESHOLD, |
95 | .down_threshold = DEF_FREQUENCY_DOWN_THRESHOLD, | |
96 | .sampling_down_factor = DEF_SAMPLING_DOWN_FACTOR, | |
97 | .ignore_nice = 0, | |
98 | .freq_step = 5, | |
b9170836 DJ |
99 | }; |
100 | ||
8e677ce8 AC |
101 | static inline cputime64_t get_cpu_idle_time_jiffy(unsigned int cpu, |
102 | cputime64_t *wall) | |
dac1c1a5 | 103 | { |
8e677ce8 AC |
104 | cputime64_t idle_time; |
105 | cputime64_t cur_wall_time; | |
106 | cputime64_t busy_time; | |
107 | ||
108 | cur_wall_time = jiffies64_to_cputime64(get_jiffies_64()); | |
109 | busy_time = cputime64_add(kstat_cpu(cpu).cpustat.user, | |
110 | kstat_cpu(cpu).cpustat.system); | |
e08f5f5b | 111 | |
8e677ce8 AC |
112 | busy_time = cputime64_add(busy_time, kstat_cpu(cpu).cpustat.irq); |
113 | busy_time = cputime64_add(busy_time, kstat_cpu(cpu).cpustat.softirq); | |
114 | busy_time = cputime64_add(busy_time, kstat_cpu(cpu).cpustat.steal); | |
115 | busy_time = cputime64_add(busy_time, kstat_cpu(cpu).cpustat.nice); | |
e08f5f5b | 116 | |
8e677ce8 AC |
117 | idle_time = cputime64_sub(cur_wall_time, busy_time); |
118 | if (wall) | |
119 | *wall = cur_wall_time; | |
e08f5f5b | 120 | |
8e677ce8 AC |
121 | return idle_time; |
122 | } | |
123 | ||
124 | static inline cputime64_t get_cpu_idle_time(unsigned int cpu, cputime64_t *wall) | |
125 | { | |
126 | u64 idle_time = get_cpu_idle_time_us(cpu, wall); | |
127 | ||
128 | if (idle_time == -1ULL) | |
129 | return get_cpu_idle_time_jiffy(cpu, wall); | |
130 | ||
131 | return idle_time; | |
dac1c1a5 DJ |
132 | } |
133 | ||
a8d7c3bc EO |
134 | /* keep track of frequency transitions */ |
135 | static int | |
136 | dbs_cpufreq_notifier(struct notifier_block *nb, unsigned long val, | |
137 | void *data) | |
138 | { | |
139 | struct cpufreq_freqs *freq = data; | |
140 | struct cpu_dbs_info_s *this_dbs_info = &per_cpu(cpu_dbs_info, | |
141 | freq->cpu); | |
142 | ||
f407a08b AC |
143 | struct cpufreq_policy *policy; |
144 | ||
26d204af PV |
145 | if (!this_dbs_info->enable) |
146 | return 0; | |
147 | ||
f407a08b AC |
148 | policy = this_dbs_info->cur_policy; |
149 | ||
150 | /* | |
151 | * we only care if our internally tracked freq moves outside | |
152 | * the 'valid' ranges of freqency available to us otherwise | |
153 | * we do not change it | |
154 | */ | |
155 | if (this_dbs_info->requested_freq > policy->max | |
156 | || this_dbs_info->requested_freq < policy->min) | |
157 | this_dbs_info->requested_freq = freq->new; | |
a8d7c3bc EO |
158 | |
159 | return 0; | |
160 | } | |
161 | ||
162 | static struct notifier_block dbs_cpufreq_notifier_block = { | |
163 | .notifier_call = dbs_cpufreq_notifier | |
164 | }; | |
165 | ||
b9170836 DJ |
166 | /************************** sysfs interface ************************/ |
167 | static ssize_t show_sampling_rate_max(struct cpufreq_policy *policy, char *buf) | |
168 | { | |
4f4d1ad6 TR |
169 | printk_once(KERN_INFO "CPUFREQ: conservative sampling_rate_max " |
170 | "sysfs file is deprecated - used by: %s\n", current->comm); | |
cef9615a | 171 | return sprintf(buf, "%u\n", -1U); |
b9170836 DJ |
172 | } |
173 | ||
174 | static ssize_t show_sampling_rate_min(struct cpufreq_policy *policy, char *buf) | |
175 | { | |
cef9615a | 176 | return sprintf(buf, "%u\n", min_sampling_rate); |
b9170836 DJ |
177 | } |
178 | ||
8e677ce8 AC |
179 | #define define_one_ro(_name) \ |
180 | static struct freq_attr _name = \ | |
b9170836 DJ |
181 | __ATTR(_name, 0444, show_##_name, NULL) |
182 | ||
183 | define_one_ro(sampling_rate_max); | |
184 | define_one_ro(sampling_rate_min); | |
185 | ||
186 | /* cpufreq_conservative Governor Tunables */ | |
187 | #define show_one(file_name, object) \ | |
188 | static ssize_t show_##file_name \ | |
189 | (struct cpufreq_policy *unused, char *buf) \ | |
190 | { \ | |
191 | return sprintf(buf, "%u\n", dbs_tuners_ins.object); \ | |
192 | } | |
193 | show_one(sampling_rate, sampling_rate); | |
194 | show_one(sampling_down_factor, sampling_down_factor); | |
195 | show_one(up_threshold, up_threshold); | |
196 | show_one(down_threshold, down_threshold); | |
001893cd | 197 | show_one(ignore_nice_load, ignore_nice); |
b9170836 DJ |
198 | show_one(freq_step, freq_step); |
199 | ||
18a7247d | 200 | static ssize_t store_sampling_down_factor(struct cpufreq_policy *unused, |
b9170836 DJ |
201 | const char *buf, size_t count) |
202 | { | |
203 | unsigned int input; | |
204 | int ret; | |
9acef487 | 205 | ret = sscanf(buf, "%u", &input); |
8e677ce8 | 206 | |
2c906b31 | 207 | if (ret != 1 || input > MAX_SAMPLING_DOWN_FACTOR || input < 1) |
b9170836 DJ |
208 | return -EINVAL; |
209 | ||
3fc54d37 | 210 | mutex_lock(&dbs_mutex); |
b9170836 | 211 | dbs_tuners_ins.sampling_down_factor = input; |
3fc54d37 | 212 | mutex_unlock(&dbs_mutex); |
b9170836 DJ |
213 | |
214 | return count; | |
215 | } | |
216 | ||
18a7247d | 217 | static ssize_t store_sampling_rate(struct cpufreq_policy *unused, |
b9170836 DJ |
218 | const char *buf, size_t count) |
219 | { | |
220 | unsigned int input; | |
221 | int ret; | |
9acef487 | 222 | ret = sscanf(buf, "%u", &input); |
b9170836 | 223 | |
8e677ce8 | 224 | if (ret != 1) |
b9170836 | 225 | return -EINVAL; |
8e677ce8 AC |
226 | |
227 | mutex_lock(&dbs_mutex); | |
cef9615a | 228 | dbs_tuners_ins.sampling_rate = max(input, min_sampling_rate); |
3fc54d37 | 229 | mutex_unlock(&dbs_mutex); |
b9170836 DJ |
230 | |
231 | return count; | |
232 | } | |
233 | ||
18a7247d | 234 | static ssize_t store_up_threshold(struct cpufreq_policy *unused, |
b9170836 DJ |
235 | const char *buf, size_t count) |
236 | { | |
237 | unsigned int input; | |
238 | int ret; | |
9acef487 | 239 | ret = sscanf(buf, "%u", &input); |
b9170836 | 240 | |
3fc54d37 | 241 | mutex_lock(&dbs_mutex); |
9acef487 | 242 | if (ret != 1 || input > 100 || |
8e677ce8 | 243 | input <= dbs_tuners_ins.down_threshold) { |
3fc54d37 | 244 | mutex_unlock(&dbs_mutex); |
b9170836 DJ |
245 | return -EINVAL; |
246 | } | |
247 | ||
248 | dbs_tuners_ins.up_threshold = input; | |
3fc54d37 | 249 | mutex_unlock(&dbs_mutex); |
b9170836 DJ |
250 | |
251 | return count; | |
252 | } | |
253 | ||
18a7247d | 254 | static ssize_t store_down_threshold(struct cpufreq_policy *unused, |
b9170836 DJ |
255 | const char *buf, size_t count) |
256 | { | |
257 | unsigned int input; | |
258 | int ret; | |
9acef487 | 259 | ret = sscanf(buf, "%u", &input); |
b9170836 | 260 | |
3fc54d37 | 261 | mutex_lock(&dbs_mutex); |
8e677ce8 AC |
262 | /* cannot be lower than 11 otherwise freq will not fall */ |
263 | if (ret != 1 || input < 11 || input > 100 || | |
264 | input >= dbs_tuners_ins.up_threshold) { | |
3fc54d37 | 265 | mutex_unlock(&dbs_mutex); |
b9170836 DJ |
266 | return -EINVAL; |
267 | } | |
268 | ||
269 | dbs_tuners_ins.down_threshold = input; | |
3fc54d37 | 270 | mutex_unlock(&dbs_mutex); |
b9170836 DJ |
271 | |
272 | return count; | |
273 | } | |
274 | ||
001893cd | 275 | static ssize_t store_ignore_nice_load(struct cpufreq_policy *policy, |
b9170836 DJ |
276 | const char *buf, size_t count) |
277 | { | |
278 | unsigned int input; | |
279 | int ret; | |
280 | ||
281 | unsigned int j; | |
18a7247d DJ |
282 | |
283 | ret = sscanf(buf, "%u", &input); | |
284 | if (ret != 1) | |
b9170836 DJ |
285 | return -EINVAL; |
286 | ||
18a7247d | 287 | if (input > 1) |
b9170836 | 288 | input = 1; |
18a7247d | 289 | |
3fc54d37 | 290 | mutex_lock(&dbs_mutex); |
18a7247d | 291 | if (input == dbs_tuners_ins.ignore_nice) { /* nothing to do */ |
3fc54d37 | 292 | mutex_unlock(&dbs_mutex); |
b9170836 DJ |
293 | return count; |
294 | } | |
295 | dbs_tuners_ins.ignore_nice = input; | |
296 | ||
8e677ce8 | 297 | /* we need to re-evaluate prev_cpu_idle */ |
dac1c1a5 | 298 | for_each_online_cpu(j) { |
8e677ce8 AC |
299 | struct cpu_dbs_info_s *dbs_info; |
300 | dbs_info = &per_cpu(cpu_dbs_info, j); | |
301 | dbs_info->prev_cpu_idle = get_cpu_idle_time(j, | |
302 | &dbs_info->prev_cpu_wall); | |
303 | if (dbs_tuners_ins.ignore_nice) | |
304 | dbs_info->prev_cpu_nice = kstat_cpu(j).cpustat.nice; | |
b9170836 | 305 | } |
3fc54d37 | 306 | mutex_unlock(&dbs_mutex); |
b9170836 DJ |
307 | |
308 | return count; | |
309 | } | |
310 | ||
311 | static ssize_t store_freq_step(struct cpufreq_policy *policy, | |
312 | const char *buf, size_t count) | |
313 | { | |
314 | unsigned int input; | |
315 | int ret; | |
18a7247d | 316 | ret = sscanf(buf, "%u", &input); |
b9170836 | 317 | |
18a7247d | 318 | if (ret != 1) |
b9170836 DJ |
319 | return -EINVAL; |
320 | ||
18a7247d | 321 | if (input > 100) |
b9170836 | 322 | input = 100; |
18a7247d | 323 | |
b9170836 DJ |
324 | /* no need to test here if freq_step is zero as the user might actually |
325 | * want this, they would be crazy though :) */ | |
3fc54d37 | 326 | mutex_lock(&dbs_mutex); |
b9170836 | 327 | dbs_tuners_ins.freq_step = input; |
3fc54d37 | 328 | mutex_unlock(&dbs_mutex); |
b9170836 DJ |
329 | |
330 | return count; | |
331 | } | |
332 | ||
333 | #define define_one_rw(_name) \ | |
334 | static struct freq_attr _name = \ | |
335 | __ATTR(_name, 0644, show_##_name, store_##_name) | |
336 | ||
337 | define_one_rw(sampling_rate); | |
338 | define_one_rw(sampling_down_factor); | |
339 | define_one_rw(up_threshold); | |
340 | define_one_rw(down_threshold); | |
001893cd | 341 | define_one_rw(ignore_nice_load); |
b9170836 DJ |
342 | define_one_rw(freq_step); |
343 | ||
9acef487 | 344 | static struct attribute *dbs_attributes[] = { |
b9170836 DJ |
345 | &sampling_rate_max.attr, |
346 | &sampling_rate_min.attr, | |
347 | &sampling_rate.attr, | |
348 | &sampling_down_factor.attr, | |
349 | &up_threshold.attr, | |
350 | &down_threshold.attr, | |
001893cd | 351 | &ignore_nice_load.attr, |
b9170836 DJ |
352 | &freq_step.attr, |
353 | NULL | |
354 | }; | |
355 | ||
356 | static struct attribute_group dbs_attr_group = { | |
357 | .attrs = dbs_attributes, | |
358 | .name = "conservative", | |
359 | }; | |
360 | ||
361 | /************************** sysfs end ************************/ | |
362 | ||
8e677ce8 | 363 | static void dbs_check_cpu(struct cpu_dbs_info_s *this_dbs_info) |
b9170836 | 364 | { |
8e677ce8 | 365 | unsigned int load = 0; |
f068c04b | 366 | unsigned int freq_target; |
b9170836 | 367 | |
8e677ce8 AC |
368 | struct cpufreq_policy *policy; |
369 | unsigned int j; | |
b9170836 | 370 | |
08a28e2e AC |
371 | policy = this_dbs_info->cur_policy; |
372 | ||
18a7247d | 373 | /* |
8e677ce8 AC |
374 | * Every sampling_rate, we check, if current idle time is less |
375 | * than 20% (default), then we try to increase frequency | |
376 | * Every sampling_rate*sampling_down_factor, we check, if current | |
377 | * idle time is more than 80%, then we try to decrease frequency | |
b9170836 | 378 | * |
18a7247d DJ |
379 | * Any frequency increase takes it to the maximum frequency. |
380 | * Frequency reduction happens at minimum steps of | |
8e677ce8 | 381 | * 5% (default) of maximum frequency |
b9170836 DJ |
382 | */ |
383 | ||
8e677ce8 AC |
384 | /* Get Absolute Load */ |
385 | for_each_cpu(j, policy->cpus) { | |
386 | struct cpu_dbs_info_s *j_dbs_info; | |
387 | cputime64_t cur_wall_time, cur_idle_time; | |
388 | unsigned int idle_time, wall_time; | |
b9170836 | 389 | |
8e677ce8 AC |
390 | j_dbs_info = &per_cpu(cpu_dbs_info, j); |
391 | ||
392 | cur_idle_time = get_cpu_idle_time(j, &cur_wall_time); | |
393 | ||
394 | wall_time = (unsigned int) cputime64_sub(cur_wall_time, | |
395 | j_dbs_info->prev_cpu_wall); | |
396 | j_dbs_info->prev_cpu_wall = cur_wall_time; | |
08a28e2e | 397 | |
8e677ce8 AC |
398 | idle_time = (unsigned int) cputime64_sub(cur_idle_time, |
399 | j_dbs_info->prev_cpu_idle); | |
400 | j_dbs_info->prev_cpu_idle = cur_idle_time; | |
b9170836 | 401 | |
8e677ce8 AC |
402 | if (dbs_tuners_ins.ignore_nice) { |
403 | cputime64_t cur_nice; | |
404 | unsigned long cur_nice_jiffies; | |
405 | ||
406 | cur_nice = cputime64_sub(kstat_cpu(j).cpustat.nice, | |
407 | j_dbs_info->prev_cpu_nice); | |
408 | /* | |
409 | * Assumption: nice time between sampling periods will | |
410 | * be less than 2^32 jiffies for 32 bit sys | |
411 | */ | |
412 | cur_nice_jiffies = (unsigned long) | |
413 | cputime64_to_jiffies64(cur_nice); | |
414 | ||
415 | j_dbs_info->prev_cpu_nice = kstat_cpu(j).cpustat.nice; | |
416 | idle_time += jiffies_to_usecs(cur_nice_jiffies); | |
417 | } | |
418 | ||
419 | if (unlikely(!wall_time || wall_time < idle_time)) | |
420 | continue; | |
421 | ||
422 | load = 100 * (wall_time - idle_time) / wall_time; | |
423 | } | |
424 | ||
425 | /* | |
426 | * break out if we 'cannot' reduce the speed as the user might | |
427 | * want freq_step to be zero | |
428 | */ | |
429 | if (dbs_tuners_ins.freq_step == 0) | |
430 | return; | |
b9170836 | 431 | |
8e677ce8 AC |
432 | /* Check for frequency increase */ |
433 | if (load > dbs_tuners_ins.up_threshold) { | |
a159b827 | 434 | this_dbs_info->down_skip = 0; |
790d76fa | 435 | |
b9170836 | 436 | /* if we are already at full speed then break out early */ |
a159b827 | 437 | if (this_dbs_info->requested_freq == policy->max) |
b9170836 | 438 | return; |
18a7247d | 439 | |
f068c04b | 440 | freq_target = (dbs_tuners_ins.freq_step * policy->max) / 100; |
b9170836 DJ |
441 | |
442 | /* max freq cannot be less than 100. But who knows.... */ | |
f068c04b DJ |
443 | if (unlikely(freq_target == 0)) |
444 | freq_target = 5; | |
18a7247d | 445 | |
f068c04b | 446 | this_dbs_info->requested_freq += freq_target; |
a159b827 AC |
447 | if (this_dbs_info->requested_freq > policy->max) |
448 | this_dbs_info->requested_freq = policy->max; | |
b9170836 | 449 | |
a159b827 | 450 | __cpufreq_driver_target(policy, this_dbs_info->requested_freq, |
b9170836 | 451 | CPUFREQ_RELATION_H); |
b9170836 DJ |
452 | return; |
453 | } | |
454 | ||
8e677ce8 AC |
455 | /* |
456 | * The optimal frequency is the frequency that is the lowest that | |
457 | * can support the current CPU usage without triggering the up | |
458 | * policy. To be safe, we focus 10 points under the threshold. | |
459 | */ | |
460 | if (load < (dbs_tuners_ins.down_threshold - 10)) { | |
f068c04b | 461 | freq_target = (dbs_tuners_ins.freq_step * policy->max) / 100; |
b9170836 | 462 | |
f068c04b | 463 | this_dbs_info->requested_freq -= freq_target; |
a159b827 AC |
464 | if (this_dbs_info->requested_freq < policy->min) |
465 | this_dbs_info->requested_freq = policy->min; | |
b9170836 | 466 | |
8e677ce8 AC |
467 | /* |
468 | * if we cannot reduce the frequency anymore, break out early | |
469 | */ | |
470 | if (policy->cur == policy->min) | |
471 | return; | |
472 | ||
a159b827 | 473 | __cpufreq_driver_target(policy, this_dbs_info->requested_freq, |
2c906b31 | 474 | CPUFREQ_RELATION_H); |
b9170836 DJ |
475 | return; |
476 | } | |
477 | } | |
478 | ||
c4028958 | 479 | static void do_dbs_timer(struct work_struct *work) |
18a7247d | 480 | { |
8e677ce8 AC |
481 | struct cpu_dbs_info_s *dbs_info = |
482 | container_of(work, struct cpu_dbs_info_s, work.work); | |
483 | unsigned int cpu = dbs_info->cpu; | |
484 | ||
485 | /* We want all CPUs to do sampling nearly on same jiffy */ | |
486 | int delay = usecs_to_jiffies(dbs_tuners_ins.sampling_rate); | |
487 | ||
488 | delay -= jiffies % delay; | |
489 | ||
ee88415c | 490 | mutex_lock(&dbs_info->timer_mutex); |
8e677ce8 AC |
491 | |
492 | dbs_check_cpu(dbs_info); | |
493 | ||
494 | queue_delayed_work_on(cpu, kconservative_wq, &dbs_info->work, delay); | |
ee88415c | 495 | mutex_unlock(&dbs_info->timer_mutex); |
18a7247d | 496 | } |
b9170836 | 497 | |
8e677ce8 | 498 | static inline void dbs_timer_init(struct cpu_dbs_info_s *dbs_info) |
b9170836 | 499 | { |
8e677ce8 AC |
500 | /* We want all CPUs to do sampling nearly on same jiffy */ |
501 | int delay = usecs_to_jiffies(dbs_tuners_ins.sampling_rate); | |
502 | delay -= jiffies % delay; | |
503 | ||
26d204af | 504 | dbs_info->enable = 1; |
8e677ce8 AC |
505 | INIT_DELAYED_WORK_DEFERRABLE(&dbs_info->work, do_dbs_timer); |
506 | queue_delayed_work_on(dbs_info->cpu, kconservative_wq, &dbs_info->work, | |
507 | delay); | |
b9170836 DJ |
508 | } |
509 | ||
8e677ce8 | 510 | static inline void dbs_timer_exit(struct cpu_dbs_info_s *dbs_info) |
b9170836 | 511 | { |
26d204af | 512 | dbs_info->enable = 0; |
b253d2b2 | 513 | cancel_delayed_work_sync(&dbs_info->work); |
b9170836 DJ |
514 | } |
515 | ||
516 | static int cpufreq_governor_dbs(struct cpufreq_policy *policy, | |
517 | unsigned int event) | |
518 | { | |
519 | unsigned int cpu = policy->cpu; | |
520 | struct cpu_dbs_info_s *this_dbs_info; | |
521 | unsigned int j; | |
914f7c31 | 522 | int rc; |
b9170836 DJ |
523 | |
524 | this_dbs_info = &per_cpu(cpu_dbs_info, cpu); | |
525 | ||
526 | switch (event) { | |
527 | case CPUFREQ_GOV_START: | |
18a7247d | 528 | if ((!cpu_online(cpu)) || (!policy->cur)) |
b9170836 DJ |
529 | return -EINVAL; |
530 | ||
3fc54d37 | 531 | mutex_lock(&dbs_mutex); |
914f7c31 JG |
532 | |
533 | rc = sysfs_create_group(&policy->kobj, &dbs_attr_group); | |
534 | if (rc) { | |
535 | mutex_unlock(&dbs_mutex); | |
536 | return rc; | |
537 | } | |
538 | ||
835481d9 | 539 | for_each_cpu(j, policy->cpus) { |
b9170836 DJ |
540 | struct cpu_dbs_info_s *j_dbs_info; |
541 | j_dbs_info = &per_cpu(cpu_dbs_info, j); | |
542 | j_dbs_info->cur_policy = policy; | |
18a7247d | 543 | |
8e677ce8 AC |
544 | j_dbs_info->prev_cpu_idle = get_cpu_idle_time(j, |
545 | &j_dbs_info->prev_cpu_wall); | |
546 | if (dbs_tuners_ins.ignore_nice) { | |
547 | j_dbs_info->prev_cpu_nice = | |
548 | kstat_cpu(j).cpustat.nice; | |
549 | } | |
b9170836 | 550 | } |
a159b827 AC |
551 | this_dbs_info->down_skip = 0; |
552 | this_dbs_info->requested_freq = policy->cur; | |
914f7c31 | 553 | |
ee88415c | 554 | mutex_init(&this_dbs_info->timer_mutex); |
b9170836 DJ |
555 | dbs_enable++; |
556 | /* | |
557 | * Start the timerschedule work, when this governor | |
558 | * is used for first time | |
559 | */ | |
560 | if (dbs_enable == 1) { | |
561 | unsigned int latency; | |
562 | /* policy latency is in nS. Convert it to uS first */ | |
2c906b31 AC |
563 | latency = policy->cpuinfo.transition_latency / 1000; |
564 | if (latency == 0) | |
565 | latency = 1; | |
b9170836 | 566 | |
cef9615a TR |
567 | /* |
568 | * conservative does not implement micro like ondemand | |
569 | * governor, thus we are bound to jiffes/HZ | |
570 | */ | |
571 | min_sampling_rate = | |
572 | MIN_SAMPLING_RATE_RATIO * jiffies_to_usecs(10); | |
573 | /* Bring kernel and HW constraints together */ | |
574 | min_sampling_rate = max(min_sampling_rate, | |
575 | MIN_LATENCY_MULTIPLIER * latency); | |
576 | dbs_tuners_ins.sampling_rate = | |
577 | max(min_sampling_rate, | |
578 | latency * LATENCY_MULTIPLIER); | |
b9170836 | 579 | |
a8d7c3bc EO |
580 | cpufreq_register_notifier( |
581 | &dbs_cpufreq_notifier_block, | |
582 | CPUFREQ_TRANSITION_NOTIFIER); | |
b9170836 | 583 | } |
3fc54d37 | 584 | mutex_unlock(&dbs_mutex); |
8e677ce8 | 585 | |
7d26e2d5 | 586 | dbs_timer_init(this_dbs_info); |
587 | ||
b9170836 DJ |
588 | break; |
589 | ||
590 | case CPUFREQ_GOV_STOP: | |
8e677ce8 | 591 | dbs_timer_exit(this_dbs_info); |
7d26e2d5 | 592 | |
593 | mutex_lock(&dbs_mutex); | |
b9170836 DJ |
594 | sysfs_remove_group(&policy->kobj, &dbs_attr_group); |
595 | dbs_enable--; | |
ee88415c | 596 | mutex_destroy(&this_dbs_info->timer_mutex); |
8e677ce8 | 597 | |
b9170836 DJ |
598 | /* |
599 | * Stop the timerschedule work, when this governor | |
600 | * is used for first time | |
601 | */ | |
8e677ce8 | 602 | if (dbs_enable == 0) |
a8d7c3bc EO |
603 | cpufreq_unregister_notifier( |
604 | &dbs_cpufreq_notifier_block, | |
605 | CPUFREQ_TRANSITION_NOTIFIER); | |
a8d7c3bc | 606 | |
3fc54d37 | 607 | mutex_unlock(&dbs_mutex); |
b9170836 DJ |
608 | |
609 | break; | |
610 | ||
611 | case CPUFREQ_GOV_LIMITS: | |
ee88415c | 612 | mutex_lock(&this_dbs_info->timer_mutex); |
b9170836 DJ |
613 | if (policy->max < this_dbs_info->cur_policy->cur) |
614 | __cpufreq_driver_target( | |
615 | this_dbs_info->cur_policy, | |
18a7247d | 616 | policy->max, CPUFREQ_RELATION_H); |
b9170836 DJ |
617 | else if (policy->min > this_dbs_info->cur_policy->cur) |
618 | __cpufreq_driver_target( | |
619 | this_dbs_info->cur_policy, | |
18a7247d | 620 | policy->min, CPUFREQ_RELATION_L); |
ee88415c | 621 | mutex_unlock(&this_dbs_info->timer_mutex); |
8e677ce8 | 622 | |
b9170836 DJ |
623 | break; |
624 | } | |
625 | return 0; | |
626 | } | |
627 | ||
c4d14bc0 SW |
628 | #ifndef CONFIG_CPU_FREQ_DEFAULT_GOV_CONSERVATIVE |
629 | static | |
630 | #endif | |
1c256245 TR |
631 | struct cpufreq_governor cpufreq_gov_conservative = { |
632 | .name = "conservative", | |
633 | .governor = cpufreq_governor_dbs, | |
634 | .max_transition_latency = TRANSITION_LATENCY_LIMIT, | |
635 | .owner = THIS_MODULE, | |
b9170836 DJ |
636 | }; |
637 | ||
638 | static int __init cpufreq_gov_dbs_init(void) | |
639 | { | |
8e677ce8 AC |
640 | int err; |
641 | ||
642 | kconservative_wq = create_workqueue("kconservative"); | |
643 | if (!kconservative_wq) { | |
644 | printk(KERN_ERR "Creation of kconservative failed\n"); | |
645 | return -EFAULT; | |
646 | } | |
647 | ||
648 | err = cpufreq_register_governor(&cpufreq_gov_conservative); | |
649 | if (err) | |
650 | destroy_workqueue(kconservative_wq); | |
651 | ||
652 | return err; | |
b9170836 DJ |
653 | } |
654 | ||
655 | static void __exit cpufreq_gov_dbs_exit(void) | |
656 | { | |
1c256245 | 657 | cpufreq_unregister_governor(&cpufreq_gov_conservative); |
8e677ce8 | 658 | destroy_workqueue(kconservative_wq); |
b9170836 DJ |
659 | } |
660 | ||
661 | ||
11a80a9c | 662 | MODULE_AUTHOR("Alexander Clouter <alex@digriz.org.uk>"); |
9acef487 | 663 | MODULE_DESCRIPTION("'cpufreq_conservative' - A dynamic cpufreq governor for " |
b9170836 DJ |
664 | "Low Latency Frequency Transition capable processors " |
665 | "optimised for use in a battery environment"); | |
9acef487 | 666 | MODULE_LICENSE("GPL"); |
b9170836 | 667 | |
6915719b JW |
668 | #ifdef CONFIG_CPU_FREQ_DEFAULT_GOV_CONSERVATIVE |
669 | fs_initcall(cpufreq_gov_dbs_init); | |
670 | #else | |
b9170836 | 671 | module_init(cpufreq_gov_dbs_init); |
6915719b | 672 | #endif |
b9170836 | 673 | module_exit(cpufreq_gov_dbs_exit); |